Abstract: To reduce latency on a shared bus during bus arbitration, a novel shared bus system uses device select lines between a bus arbiter and the bus devices to select the bus slave concurrently with the granting of the shared bus to the bus master. Specifically, a bus device requests the use of the shared bus by driving an active state on a bus request terminal and driving a destination ID value corresponding to the desired bus slave to the bus arbiter. The bus arbiter then drives an active state on a bus grant output terminal coupled to the bus grant input terminal of the requesting device. Concurrently, the bus arbiter drives an active state on the device select output terminal coupled to the device select input terminal of the desired bus slave. In addition posted read request tagging can be simplified using a transaction ID bus to supplement the shared bus.

Abstract: A DMA controller has a first data buffer and a second data buffer. First data from a first bus can be loaded into the first data buffer at the same time that second data from a second bus is loaded into the second data buffer. Once the data is present in the first and second data buffers, the first data in the first data buffer can be supplied to the second bus at the same time that the second data in the second data buffer is supplied to the first bus (or alternatively to a third bus). In some embodiments, the second bus is a high speed parallel bus and the first and third data buses are I/O data buses for coupling the DMA controller to codecs. In some embodiments, data from the first data bus can be loaded into the first data buffer at the same time that data in the second data buffer is supplied to the third data bus. An address generator generates addresses onto the second data bus for the transfer of data between the DMA controller and the second data bus.

Abstract: To reduce latency for priority requests in a bus system, non-priority requests to a bus device are separated from priority requests so that the priority requests can be processed before the non-priority requests. Priority requests are transferred over a priority request bus. Standard non-priority requests are sent over a shared bus coupled to several bus devices. One of the bus devices contains a request buffer to receive non-priority requests from the shared bus and a priority buffer to receive priority requests from the priority request bus. Requests in the priority buffer are processed before requests in the request buffer. In one embodiment, a multiplexer having input ports coupled to the request buffer and the priority buffer is configured so that the multiplexer outputs priority requests if there are priority requests in the priority buffer.

Abstract: A FIFO storage circuit stores data transferred over a data bus in data groups having of one or more data units. The FIFO buffer includes a number of storage locations each configured to store a single unit of data. Each data groups on the data bus is accompanied by data-size information. A storage-location availability decoder receives the data-size information and allocates a number of consecutive storage locations in the FIFO storage circuit to allow for storage of the incoming data group. The FIFO then stores each unit of the data group, along with a data tag common to each unit, in consecutive storage locations. The next stored data group is then assigned a new data tag. The FIFO storage circuit reads from the storage locations in order of data tag. Checking consecutive storage locations for equivalent data tags enables the FIFO storage circuit to output the appropriate number of units for each data group.

Abstract: A DMA controller receives size data in association with a DMA request. If first size data is received, a first amount of data (for example, one word) is transferred through the DMA controller for the DMA request. If, on the other hand, second size data is received, then a second amount of data (for example, two words) is transferred through the DMA controller for the DMA request. In the event that a DMA request cannot be serviced when received, the DMA request is stored in the DMA controller for later servicing. Size data for a DMA request is stored so that the size of the data transfer will be known when the stored DMA request is serviced. Using this size data, a single DMA channel can support data transfers of different sizes. In some embodiments, size data is used to increment a DMA current address register by the correct amount after the data associated with the size data is transferred through the DMA controller.

Abstract: A DMA controller receives a DMA request from a DMA channel and generates a transaction request ID identifying the requested transaction as well as a codec ID identifying the DMA channel. The codec ID is stored in the DMA controller in association with the transaction request ID. The DMA controller obtains control of the bus and outputs a transaction request onto the bus along with the transaction request ID. The DMA controller then relinquishes control of the bus. A device on the bus returns the transaction request ID when it responds (the transaction request ID is "tagged" to the response). The DMA controller uses the returned transaction request ID to look up the codec ID stored in the DMA controller in association with that transaction request ID. The DMA controller uses the codec ID to associate the response with the correct DMA channel.

Abstract: A multi-standard start code detector system receives an incoming compressed video bit stream. The compressed video bit stream includes start codes interspersed between variable length bit streams. Assuming error-free useful data, zero stuffing bits preceding a next start code are rapidly discarded on a bit group-wise basis. A group of current bits are processed by comparing them to a start code feature pattern. If a match is not detected, the entire group of bits is concurrently replaced by a next group. However, because the zero stuffing bits are of variable length, the exact position of the start code pattern is unknown. Preceding each replacement operation, the next group of bits is processed to predetect a start code feature. Upon predetecting a start code feature, the replacement operation is modified to ensure that a group of current data bits will include the start code feature pattern. This allows the number of searched for patterns to be reduced and streamlines the comparison operation.

Abstract: A PCI interface includes a PCI core that operates at the PCI bus frequency and glue logic which provides an interface to a higher frequency clock domain. The glue logic includes FIFO buffers for addresses and data coming from or going to the PCI bus, and synchronizers for control signals coming from or going to the PCI core. In one embodiment, a novel synchronizer includes three flip-flops at least two of which are JK flip-flops.

Abstract: A method and system for simultaneous retrieval of snoop address information in conjunction with the retrieval/storing of a cache line load/store operation. The method and system are implemented in a data processing system comprising at least one processor having an integrated controller, a cache external to the at least one processor, and an interface between the at least one processor and the external cache. The external cache includes a tag array and a data array. Standard synchronous static Random Access Memory (RAM) is used for the tag array, while synchronous burst made static RAM is used for the data array. The interface includes a shared address bus, a load address connection and an increment address connection. A cache line load/store operation is executed by placing an address for the operation on the shared address bus, and latching the address with the external cache using a signal from the load address connection.

Abstract: A system and method for converting data with one error correction code format to data with a different error correction code format, including provisions for correcting the errors in the input data and checkbits. The invention reduces the levels of logic needed to accomplish the conversion. In one form, checkbits for the converted data are generated from the input data at the same time that the input data correction syndromes are generated from the input data and checkbits. Multiple sets of error pointers are then used to simultaneously correct both the converted data and the converted data checkbits.